API and MCP

Every number on this site is a static JSON file. No key, no rate limit, cache as you like. Every payload carries generated, licence and base. Code, examples and the MCP server: github.com/Infer-CPI/api.

Base URL

https://infercpi.com/v1/

Endpoints

File	What	Updated
`latest.json`	Last value of every series: `latest["price_index/chained_price_idx"] → {ts, date, value}`, plus the latest editorial note	daily
`index.json`	Catalogue: every series with point count, first and last date, method string and URL	daily
`series/<signal>/<key>.json`	Full history as `points: [[unix_ts, value], …]` plus `method`	with its signal
`basket.json`	Every model priced this week: key, provider, input, output and blended $/1M, context, Epoch ECI, open or closed weights	weekly
`events.json`	Launches, price cuts, price increases and removals in the labs' own price lists since Sep 2023, with old and new price	daily
`speed.json`	Our own daily speed run: time to first token and output tokens per second, per model and route	daily
`token_mix.json`	Tokens routed through OpenRouter per week, by lab and by region (trillions)	weekly
`capex.json`	Five payers' quarterly capex, chip and memory makers' revenue, cloud segment revenue, routed tokens and paid price by quarter	weekly
`spread_models.json`	Cheapest and dearest scored model each week	weekly
`notes.json`	The human-written weekly lines, newest first	weekly
`names.json`	Display names for the model keys used in series names	daily

Series families

Use them as series/<signal>/<key>.json. The full list of keys is in index.json.

Signal	What it holds	Method
`price_index`	Same-model chained index, composition index, quality-adjusted line, price spread, cheapest price at each capability level, open vs closed, host discounts, DeepSeek floor lift	chained · open vs closed
`weighted_index`	What buyers pay (usage-weighted $/1M) and the Törnqvist index	paid
`or_rankings`	OpenRouter token shares: closed US labs, Chinese labs	volume
`implied_margin`	Serving-cost band and implied gross margin for closed, open and paid prices	margin
`self_host`	Desk boxes and H100s (rented by source, owned) in $/1M against the API, break-even, price to compete	leak
`task_cost`	Dollars and tokens per completed task, reasoning share, accuracy, per model and median	cost per task
`gpu_spot`, `gpu_neocloud`, `aws_gpu`	GPU rent per hour: Vast.ai marketplace, RunPod, CoreWeave, Lambda, Nebius, AWS	leak
`hf_downloads`	Hugging Face 30-day downloads: local-format (GGUF) repos and the open-weight labs' own	leak
`hyperscaler_capex`, `seller_revenue`, `cloud_revenue`, `oracle_credit`, `nvda_oblig`	SEC filings: capex, chip and memory revenue, cloud segment revenue, Oracle backlog, NVIDIA purchase obligations	who pays
`speed`	Daily median time to first token and tokens per second	speed

Examples

curl -s https://infercpi.com/v1/latest.json | jq '.latest["weighted_index/paid_usd_1m"]'
curl -s https://infercpi.com/v1/series/price_index/chained_price_idx.json | jq '.points[-1]'

import json, urllib.request
get = lambda p: json.load(urllib.request.urlopen(f"https://infercpi.com/v1/{p}"))
latest = get("latest.json")["latest"]
print("buyers pay", latest["weighted_index/paid_usd_1m"]["value"], "$/1M")
for ts, v in get("series/price_index/cheapest_eci145_usd_1m.json")["points"][-4:]:
    print(ts, v)

MCP server for agents

Ten read-only tools over the same files: index_latest, index_series, index_catalogue, basket, model_price, events, speed, open_vs_closed, self_hosting, cost_per_task. One file, one dependency.

mkdir -p ~/.local/share/infercpi
curl -so ~/.local/share/infercpi/mcp_server.py https://raw.githubusercontent.com/Infer-CPI/api/main/mcp_server.py
pip install mcp                     # mcp 1.x or 2.x both work
claude mcp add infercpi -s user -- python3 ~/.local/share/infercpi/mcp_server.py   # Claude Code

{ "mcpServers": { "infercpi": { "command": "python3", "args": ["/home/YOU/.local/share/infercpi/mcp_server.py"] } } }   // Claude Desktop

A hosted endpoint is coming; until then the server runs locally and reads infercpi.com.

Licence and citation

Index values are CC BY 4.0: use them commercially or not, with credit. Cite as InferCPI, infercpi.com. Inputs keep their own terms: LiteLLM and pydantic/genai-prices (MIT), Epoch AI Capabilities Index (CC BY 4.0), OpenRouter rankings (CC BY 4.0), SEC filings (public domain). Speed and cost per task are our own measurements.